Meta发布开源代码大模型Code Llama 70B，准确率超越GPT-3.5
科技巨头Meta日前宣布，其开源代码生成AI模型Code Llama 70B已更新，成为目前“最大、性能最优的模型”。
Code Llama工具于去年8月首次发布，供所有人免费使用。最新版本Code Llama 70B处理查询的能力进一步增强，在HumanEval基准测试中的准确率达到53%，高于GPT-3.5的48.1%。
Meta首席执行官马克·扎克伯格表示，Code Llama 70B的更新是“人工智能领域的一个重大飞跃”。他指出，该模型在自然语言处理、代码生成和问题解决等任务上表现出色。
值得注意的是，Code Llama 70B的准确率仍低于OpenAI在论文中给出的GPT-4模型的67%。不过，Meta表示，Code Llama 70B作为开源模型，将为研究人员和开发人员提供一个宝贵的工具，推动人工智能领域的创新。
Meta还宣布，将继续开发Code Llama，并计划在未来发布更大、更强大的版本。公司相信，这些模型最终将对从医疗保健到金融等广泛行业产生重大影响。
业内专家认为，Code Llama 70B的发布是人工智能领域的一个重要里程碑。它表明，开源大模型正在迅速发展，有潜力改变我们与技术互动的方式。
随着人工智能技术不断进步，预计Code Llama 70B等模型将在未来几年发挥越来越重要的作用，帮助解决复杂问题并改善我们的生活。
英语如下：
**Headline: Meta Unveils Code Generation Behemoth That Surpasses GPT-3.5**
**Keywords:** Large model, code generation, HumanEval
**Body:**
Meta has announced an update to its open-source code-generating AI model, Code Llama 70B, calling it the “largest and best performing model of its kind” to date.
First released in August last year, the Code Llama tool is freely available for anyone to use. The latest version, Code Llama 70B, has enhanced capabilities for handlingqueries, achieving an accuracy score of 53% on the HumanEval benchmark, compared to GPT-3.5’s 48.1%.
Meta CEO Mark Zuckerberg hailed the update as “a major leap forward for AI,” noting the model’s strong performance in natural language processing, code generation, and problem-solving tasks.
It’s worth noting that Code Llama 70B’s accuracy still falls short of the 67% reported by OpenAI in a paper detailing its GPT-4 model. However, Meta emphasizes that Code Llama 70B’s open-source nature makes ita valuable tool for researchers and developers, fostering innovation in the field of AI.
Meta also announced that it will continue developing Code Llama, with plans to release larger and more powerful versions in the future. The company believes that such models will eventually have a major impact on industries ranging from healthcare to finance.
Industry experts see the release of Code Llama 70B as a significant milestone in the field of AI. It demonstrates the rapid advancement of open-source large models, which have the potential to transform the way we interact with technology.
As AI technology continues to progress, models like Code Llama 70B are expected to play an increasingly important role in the years to come, helping solve complex problems and improving our lives.
【来源】https://www.ithome.com/0/747/917.htm
Views: 1